One Country Can Block a Killer-Robot Treaty. Right Now, Two of Them Are Trying To.

 

The rule governing these talks is simple and, for anyone hoping they succeed, brutal: the Convention on Certain Conventional Weapons operates entirely by consensus. All 128 member states have to agree, which means any single one of them can kill the deal by saying no. Right now, according to diplomats speaking to Reuters from this week's Geneva session, two of the most consequential militaries on earth are doing exactly that.

The document at stake is a non-binding text that wouldn't itself ban anything. Its job is narrower and more procedural: opening the door to actual negotiations on rules for Lethal Autonomous Weapons Systems, drones, targeting software, and other AI-enabled weapons capable of identifying and engaging targets with varying degrees of human involvement. A rolling draft the broader working group has developed over several years takes a two-tier approach, an outright ban on systems that can't be operated under meaningful human control, and binding regulation for everything else that can. More than 70 states endorsed that framework back in March. It's meant to be the floor everyone builds from here.

The United States and Russia are pushing back on that floor from the same general direction, though not for identical reasons voiced in public. Both governments argue existing international humanitarian law already covers how these weapons should be used in war, making a new treaty premature or unnecessary. One senior diplomat told Reuters that Washington's position has specifically hardened over the past six months, moving toward something closer to voluntary guidelines rather than a legally binding instrument. That's a real shift in emphasis, not a change in the basic argument, and it's worth noting Washington isn't purely obstructing behind closed doors: at least six countries received formal diplomatic outreach from the US specifically trying to build support for its preferred version of the text, which one source characterized as genuine engagement rather than pure stonewalling. The dispute, per that reporting, comes down to specific words, including how the text defines "lethality" and what it says about preserving human judgment and control.

The clock matters here more than the substance of any single disputed phrase. The group's current negotiating mandate runs out in November, and Friday was supposed to be the moment diplomats either lock in consensus language or watch years of drafting work potentially evaporate. "The window is getting smaller to getting an agreement," one diplomatic source said. Another was blunter about what's actually at risk: key provisions debated for years could simply get dropped in a last-minute scramble to produce anything at all, or the whole effort could end without a deal.

It's worth being precise about who's actually opposed here, because the US and Russia aren't isolated. Reporting on the broader diplomatic landscape names Australia, India, Israel, Japan, South Korea, and the United Kingdom as also resisting legally binding regulation, a group spanning genuinely different political systems and strategic postures, united mainly by being countries that either build or rely heavily on advanced military technology. Brazil, Ireland, and a bloc of African states have been the most consistent advocates for a binding treaty with real human-control safeguards. That split tracks a fairly predictable line: countries investing the most in autonomous weapons development tend to want maximum flexibility in how they're eventually allowed to use them, while countries without that same technological stake see less cost in supporting firm limits.

UN Secretary-General António Guterres and Red Cross President Mirjana Spoljaric didn't wait for Friday's outcome before sounding an alarm. In August, both warned the world was approaching what they called a moral red line, a threshold where letting machines make life-and-death targeting decisions with diminishing human oversight becomes normalized simply because no binding rule ever arrived in time to stop it. Whether Friday's session produced a deal or joined the list of near-misses this process has already accumulated, the underlying tension isn't going anywhere: the same handful of states most capable of building these weapons remain the ones with the least appetite for being told, by treaty, how they're allowed to use them.

The Air Force's Answer to Losing a Quarter of Its Drone Fleet: Stop Building Drones That Are Too Good to Lose

 

"Although we have plenty of MQ-9As today, we are going through them at a rate that is concerning to us." That's Lt. Gen. Christopher Niemi, the Air Force's chief modernization officer, describing why a program originally planned to reach initial capability by 2031 just got compressed to as early as 2029. Since the Iran war began, the Air Force has lost at least 45 MQ-9 Reapers, roughly a quarter of its entire fleet, and the service's own math on that loss rate is blunt: it can't keep losing $50 million drones at this pace and expect the fleet to hold up.

What's notable is what the Air Force decided not to do about it. Officials looked seriously at General Atomics' MQ-9B, a bigger, more advanced, more expensive version of the current Reaper, and turned it down. Niemi's reasoning cuts right to the actual problem: "If you don't like losing an MQ-9A that cost on the order of $50 million, with the sensors, you're probably not going to like losing an MQ-9B at a higher price point." A more capable replacement doesn't solve an attrition problem, it just makes each loss more expensive. Instead, the Air Force is going all-in on Massed Modular Aircraft, a program explicitly designed around the idea that some of these drones are going to get shot down, and that's fine, as long as replacing them is cheap and fast.

The target price is $10 million per aircraft, a fifth of what an MQ-9A with sensors costs today. To hit that number, the Air Force is deliberately choosing to build something less impressive rather than more. Niemi described the design philosophy as "80 percent is good enough," using mature, already-proven technology instead of pushing for cutting-edge performance. MMA drones will look a lot like current Reapers, long, efficient wings and turboprop engines, because that's a well-understood, inexpensive design rather than something requiring years of exotic engineering. The one place the Air Force is willing to spend real effort is autonomy: more automated flight control, letting a single Airman fly multiple drones instead of one pilot per aircraft, cutting personnel costs even as the hardware itself stays deliberately unambitious.

The contrast Niemi drew with the F-35 is the clearest window into why this approach matters beyond just MMA specifically. Asked about production flexibility, he didn't talk about MMA's own numbers, he talked about what happens when a fighter this complex needs to scale up fast. "If you wanted to suddenly ramp up production of something like an F-35, it's not a limiting factor so much as money, as the ability to build out your production capacity," he said. "In major combat operations, depending on who you're fighting, we could be losing dozens a day, and so you have to figure out how do you adjust the industrial capacity." That's an Air Force official openly stating that America's most advanced fighter jet cannot be surged fast enough to replace losses in a genuinely intense war, a real admission buried inside a briefing that was ostensibly about a drone program.

MMA sits alongside Collaborative Combat Aircraft, the Air Force's other big autonomous drone effort, and the differences between the two are deliberate rather than incidental. CCA is built to look and act more like a fighter, roughly $20 million a unit, flying combat missions with a higher degree of autonomy. MMA is closer to a delivery truck, in Niemi's own words, carrying a sensor package or a weapon to where it needs to be rather than hunting and engaging targets on its own, with roughly twice the range of a CCA but at slower speed. The Air Force wants at least 180 of them, both programs sharing open, standardized architecture specifically so new sensors or weapons from any vendor can be swapped in quickly, the same "conforms to this interface, we can do it speedily" flexibility Niemi contrasted directly against the F-35's notoriously difficult upgrade path.

General Atomics, whose MQ-9B got passed over, isn't sitting this one out. The company is pitching a new design called Wildfire specifically for the MMA competition, with a spokesman claiming it can deliver "years earlier" than competitors while keeping "everything that makes Reaper special" at a price that "doesn't break the bank." Whether that promise holds up against a 2029 target the Air Force itself only recently decided to accelerate is the real test ahead, one further complicated by exactly the kind of losses that made this timeline urgent in the first place.